Towards a Robust Method for Estimating τ Using 21 cm Data and Machine Learning

The Epoch of Reionization (EoR) is a period after the Dark Ages where photons produced by the first stars began to break the neutral hydrogen atoms in the intergalactic medium (IGM) apart. This free electron content creates an opacity to cosmic microwave background (CMB) photons which is summarized in integrated optical depth to reionization τ (tau). Uncertainty in τ leads to uncertainty in cosmological parameters extracted from the CMB; therefore, methods of directly inferring it are highly valuable. τ, as a summary of the ionization fraction as a function of time, is in principle directly measured by 21 cm brightness temperature measurements, e.g., future observations from the Hydrogen Epoch of Reionization Array (HERA), because these measurements are directly sensitive to the fraction of space that is ionized or neutral hydrogen and can measure this at different redshifts during the reionization process. Directly extracting τ from 21 cm measurements requires an analysis approach that uses all the information in the images produced by HERA, and to test these algorithms requires simulations. To find a robust method to estimate τ, we want to make our estimator as independent as possible of the uncertain aspects of current simulations. Therefore, we chose to consider simulations from two different simulators—21cmfast and Zreion—to characterize their similarities and differences. Zreion uses a biased density field prescription to create the ionization field (Battaglia et al. 2013; ApJ 776 31), whereas 21cmfast (Murray et al. 2020; JOSS 5(54) 2582; Mesinger et al. 2011; MNRAS 411 955) uses a more physical description of the star formation and ionized photon creation process. They produce different morphologies of ionized and neutral regions. They also can produce different ionization histories, with respect to the duration and timing of reionization. We compare the different morphologies by looking at the power spectrum and the different ionization histories by the midpoint, width, and asymmetry parameters of the ionization history and find common parameters where each simulator produces similar ionization histories. We are currently working on training machine learning algorithms on both the simulations to study how they perform with different morphologies, yet the same underlying density field, cosmological, and identical ionization histories.